BEGIN:VCALENDAR
VERSION:2.0
PRODID:icalendar-ruby
CALSCALE:GREGORIAN
X-WR-CALNAME:CAM Colloquium - Preston Culbertson\, Department of Computer S
 cience\, Cornell University
X-WR-TIMEZONE:Eastern Time (US & Canada)
BEGIN:VEVENT
DTSTAMP:20260518T160123Z
UID:tag:localist.com\,2008:EventInstance_52277300842808
DTSTART:20260320T194500Z
DTEND:20260320T204500Z
DESCRIPTION:Title: Stochastic Control Barrier Functions for Robot Safety\n\
 nAbstract: Control barrier functions certify safety for deterministic syst
 ems by rendering a set of safe states forward invariant\, but these guaran
 tees are only as strong as the underlying model. For contact-rich robotics
  tasks like manipulation and locomotion\, unmodeled dynamics and stochasti
 c disturbances render such certificates unreliable. Rather than seeking mo
 re accurate deterministic models\, I'll discuss how classical martingale t
 echniques for discrete-time stochastic stability (Kushner\, 1967) can be a
 dapted to construct probabilistic forward invariance guarantees that expli
 citly account for model uncertainty.\n\nThe approach enforces a supermarti
 ngale condition on barrier-like functions\, yielding bounds on the probabi
 lity of constraint violation under state-dependent stochastic disturbances
 . Since real disturbance distributions are typically unknown and non-Gauss
 ian\, we learn them from data using deep generative models\, then synthesi
 ze controllers under the learned stochasticity. Real-world results on quad
 rotor flight with unmodeled payloads and humanoid locomotion demonstrate t
 hat explicitly modeling and accounting for model error can produce more ro
 bust behavior in practice.\n\nBio: Preston Culbertson is an Assistant Prof
 essor of Computer Science at Cornell. He leads the Praxis Lab\, where they
  study how robots can learn to be robust and reliable in the real world.\n
 \nHe wants to understand robustness in robot learning. His work integrates
  machine learning\, numerical optimization\, and control theory to study h
 ow robots can remain reliable when models\, sensors\, or hardware are impe
 rfect. The goal is to develop robotic systems that can manage uncertainty\
 , adapt\, and improvise when deployed in messy real-world environments.
GEO:42.443451;-76.481506
LOCATION:Frank H. T. Rhodes Hall\, 655
SUMMARY:CAM Colloquium - Preston Culbertson\, Department of Computer Scienc
 e\, Cornell University
URL;VALUE=URI:https://events.cornell.edu/event/cam-colloquium-preston-culbe
 rtson-department-of-computer-science-cornell-university
CATEGORIES:Colloquium
END:VEVENT
END:VCALENDAR